Skip to content
Publications

Dataset · Zenodo · June 12, 2026

ArchiveGene Corpus (v1.0): Uzbek Historical Archival Text Benchmark for NER, Coreference, and Genealogical Relation Extraction

Dauletov Adilbek Yusupbayevich, Noila Matyakubova, Ikrom Islomov

Abstract

ArchiveGene is a synthetic, multi-layer annotated corpus for information extraction from Uzbek historical archival texts. The corpus contains 1,000 documents split into training (700), validation (150), and test (150) sets. Each document is annotated across five layers: named entities, person mentions, coreference clusters, genealogical relation triples, and final structured tuples. The corpus is generated by a deterministic, template-based pipeline, enabling full reproducibility. It is designed as a benchmark for Named Entity Recognition, Coreference Resolution, and Genealogical Relation Extraction tasks. Format: JSON-L (one document per line). See README.md for the full schema and annotation guidelines.v1.1: Added UNCERTAIN mention type (8th type); extended document text.

Topic ModelingNatural Language Processing TechniquesDigital Humanities and Scholarshipnamed entity recognitioncoreference resolutioninformation extraction

More by Noila Matyakubova